Translate array drop glue using MIR
I was a bit lazy here and used a usize-based index instead of a pointer iteration. Do you think this is important @eddyb?
r? @eddyb
The main changes around rustc::ty::Layout::struct and rustc_trans:adt:
* Added primitive_align field which stores alignment before repr align
* Always emit field padding when generating the LLVM struct fields
* Added methods for adjusting field indexes from the layout index to the
LLVM struct field index
The main user of this information is rustc_trans::adt::struct_llfields
which determines the LLVM fields to be used by LLVM, including padding
fields.
Exclude top-level macro expansions from source location override.
It occurred to me that a simple heuristic can address the issue #36382: any macros that expand into items (including `include!()`) don't need to be stepped over because there's not code to step through above a function scope level.
r? @michaelwoerister
According to the LLVM reference:
> A value of 0 or an omitted align argument means that the operation has
the ABI alignment for the target.
So loads/stores of fields of packed structs need to have their align set
to 1. Implement that by tracking the alignment of `LvalueRef`s.
Fixes#39376.
LLVM Core C bindings provide this function for all the versions back to what we support (3.7), and
helps to avoid this unnecessary builder->function transition every time. Also a negative diff.
Fix transmute::<T, U> where T requires a bigger alignment than U
For transmute::<T, U> we simply pointercast the destination from a U
pointer to a T pointer, without providing any alignment information,
thus LLVM assumes that the destination is aligned to hold a value of
type T, which is not necessarily true. This can lead to LLVM emitting
machine instructions that assume said alignment, and thus cause aborts.
To fix this, we need to provide the actual alignment to store_operand()
and in turn to store() so they can set the proper alignment information
on the stores and LLVM can emit the proper machine instructions.
Fixes#32947
For transmute::<T, U> we simply pointercast the destination from a U
pointer to a T pointer, without providing any alignment information,
thus LLVM assumes that the destination is aligned to hold a value of
type T, which is not necessarily true. This can lead to LLVM emitting
machine instructions that assume said alignment, and thus cause aborts.
To fix this, we need to provide the actual alignment to store_operand()
and in turn to store() so they can set the proper alignment information
on the stores and LLVM can emit the proper machine instructions.
Fixes#32947
Since discriminants do not support i128 yet, lets just calculate the boundaries within the 64 bits
that are supported. This also avoids an issue with bootstrapping on 32 bit systems due to #38727.
